Skip to content

pixi upgrade do not follow pypi dependencies index #4659

@Jiacheng-WU

Description

@Jiacheng-WU

Checks

  • I have checked that this issue has not already been reported.

  • I have confirmed this bug exists on the latest version of pixi, using pixi --version.

Reproducible example

Example pixi.toml

[workspace]
authors = [""]
channels = ["conda-forge"]
name = "test"
platforms = ["linux-64"]
version = "0.1.0"

[tasks]

[dependencies]
python = ">=3.13.7,<3.14"

[pypi-dependencies]
triton = { version = ">=3.4.0, <4", index = "https://pypi.org/simple" }
onnx = { version = ">=1.19.0, <2", index = "https://pypi.org/simple" }
pytorch-triton = { version = "*", index = "https://download.pytorch.org/whl/nightly/cu130" }
torch = { version = "*", index = "https://download.pytorch.org/whl/nightly/cu130" }
torchvision = { version = "*", index = "https://download.pytorch.org/whl/nightly/cu130" }

Example shell commands

pixi install # perfectly successful, and do use torch recent version downloaded from cu130 nightly
pixi upgrade # it then will update torch from pypi.org with oldest version, and even rewrite version="*"

Issue description

During pixi upgrade, it seems somehow follow the pypi, rather than follow the "index" provided inside [pypi-dependecies], which seems wired. Or does pixi think stable version (2.8.0) is newer than latest version (2.10.0.devXXXXXXXX)?

I am not so sure about this case, but perhaps I think upgrade should not downgrade the nightly version to stable version?

Besides, I have tried to add "extra-index-urls" in [pypi-options] sections, however, since the "triton" simultaneously happened in both "pypi.org" and "download.pytorch.org", and then pixi just find that in "download.pytorch.org", which also not follows the index, even though I tried different "index-strategy".

Expected behavior

Pixi may should consider the "index" in [pypi-dependencies] as the first priority consideration.
The nightly version should be latest, so pixi upgrade should not revert the version of nightly torch version.

Metadata

Metadata

Assignees

Labels

bugSomething isn't workingpypiIssue related to PyPI dependencies

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions